For MJ ; code for MH error

[reubenpjacob]

hi MJ
i post the code and data for the model. it is a large 2-country model .

but i am estimating only 8 ( +4) parameters including the shock AR(1) s and std errors.
i am only using 2 data series and 2 shocks. i have a lot of data with me, but the optimisation got stuck and hence i thought of using minimal data , get reasonable estimates and then add more data. but even now, i cannot get a proper mode with these gradient methods. i will need to code something myself to get the mode, i think.

i did the MCMC without getting the mode, for the moment.
that may be the reason that it is getting me this 'warning' sign.

anyways, thanks for your help and please examine the files.

cheers
reuben

[MichelJuillard]

We couldn't replicate the singular matrix warning on line 68 of
DsgeLikelihood1
I gather that it is a very exceptional message.
On the other hand, I remarked that the initial value of objective function to be minimized is extremely large, indicating major inconsistency between the model under the prior mean and the data.
You should calibrate your model with the prior mean and run stoch_simul, then compare the theoretical moments and the corresponding moments in the data. Maybe you will spot something odd that will indicate you where is the problem.
I don't think that finding a better optimizing strategy will cure it.

Kind regards

Michel

[reubenpjacob]

thanks michel. i will try out that strategy.